A self-organizing neural network for learning and generating sequences of target-directed movements in the context of a delta-lognormal synergy

This paper shows how a high level neural network can exploit the basic knowledge that emerges from the delta-lognormal theory to learn and control the generation of sequences of target-directed movements. The neural network is a topology preserving map representing the external working space and composed by a grid of leaky integrators simulating neurons. If the input vector /spl Xi/(t) represents the external end-point movement (the pen-tip track during handwriting for example) then, the global activation of the map, that is a sort of competitive population coding, is strictly correlated with the kinematic state of the ongoing external movement. In this context it is possible to detect the synchronization instant between two consecutive motor strokes and finally control both the generation and learning of a sequences of target-directed movements.